energies

Article

Application of the Extension Taguchi Method toOptimal Capability Planning of a Stand-alonePower System

Meng-Hui Wang 1, Mei-Ling Huang 2,*, Zi-Yi Zhan 1 and Chong-Jie Huang 1

1 Department of Electrical Engineering, National Chin-Yi University of Technology, No. 57, Sec. 2,Chung-Shan Rd., Taiping District, Taichung 41170, Taiwan; wangmh@ncut.edu.tw (M.-H.W.);cstd33@gmail.com (Z.-Y.Z.); 4a212110@gmail.com (C.-J.H.)

2 Department of Industrial Engineering and Management, National Chin-Yi University of Technology, No. 57,Sec. 2, Chung-Shan Rd., Taiping District, Taichung 41170, Taiwan

* Correspondence: huangml@ncut.edu.tw; Tel.: +886-4-2392-4505 (ext. 7653)

Academic Editor: Mark DeinertReceived: 17 October 2015; Accepted: 26 February 2016; Published: 8 March 2016

Abstract: An Extension Taguchi Method (ETM) is proposed on the optimized allocation of equipmentcapacity for solar cell power generation, wind power generation, full cells, electrolyzer and hydrogentanks. The ETM is based on the domain knowledge containing the product specifications andallocation levels provided by suppliers and design factors since most of the renewable energyequipment available in the market comes with a specific capacity. A proper orthogonal array isused to collect 18 sets of simulation responses. The extension theory is introduced to determinethe correlation function, and factor effects are used to identify the optimized capacity allocation.The hours of power shortage are simulated using Matlab for all capacity allocations at the lowestestablishment cost and the optimized capacity allocation of loss of load probability (LOLP). Finally, theextension theory, extension AHP theory, ETM and Analytic Hierarchy Process (AHP) are used todetermine the optimized capacity allocation of the system. Results are compared for the above fouroptimization simulation methods and verify that the proposed ETM surpasses the others on achievingthe optimized capacity allocation.

Keywords: stand-alone power system (SAPS); extension theory; extension Taguchi method (ETM);analytic hierarchy process (AHP); loss of load probability (LOLP)

1. Introduction

This paper integrates the renewable energy equipment of the photovoltaic (PV) powersystem, wind power, fuel cell, electrolyzer and hydrogen tank into one stand-alone power system.The optimized capacity allocation is an important issue for a stand-alone renewable energy system.A stand-alone renewable energy system is usually installed in remote mountains, on an island orwhere the utility power grid does not reach. A poor capacity allocation will not balance the powersupply and demand and waste the costs for the system establishment. The simulated analysis shallbe the reference of capability configuration [1,2]. The allocation of system device capacity affects thestand-alone power system [3,4]. When the device capability configuration is not appropriate, it maycause power shortages, or generate too much electricity while the system is unavailable, which wastesmoney [5,6]. On the contrary, with reasonable system allocation, the operation of power supply systemwill be set at the optimum cost.

Nelson et al. applied particle swarm optimization on the optimal sizing of a stand-alone wind/fuelcell power system [7]. They presented the unit sizing and cost analysis of stand-alone hybrid

Energies 2016, 9, 174; doi:10.3390/en9030174 www.mdpi.com/journal/energies

Energies 2016, 9, 174 2 of 17

wind/PV/fuel cell power generation systems. Ekren et al. [8] used simulated annealing method for thesize optimization of a PV/wind hybrid energy conversion system with battery storage. The drawbackof this method is that it takes too much computing time and the computing results needs to be adjustedthrough experience to fit in the specifications of the common practical products.

Yang et al. [9] applied genetic algorithm to calculate the optimal system configuration withrelative computational simplicity. Some commercial software like HOMER [10], IHOGA [11] weredeveloped to figure out the optimal allocation. Based on the simulated weather data, this studyestimates the possibility of loss of load probability (LOLP) for one or more years. Taguchi qualityengineering was developed and advocated by Dr. Taguchi Genichi [12] in 1950. The orthogonalarray experiment design and analysis of variance (ANOVA) were used for the analysis with a limitedexperimental data to effectively improve product quality [12]. The Taguchi method is widely appliedin the manufacturing field, but this study is the first to apply the Taguchi method in the optimizationof hybrid power systems.

This paper combines the extension theory with the Taguchi method, selects the proper designfactors and levels via the orthogonal array to greatly decrease the number of experiments to acquirethe optimum experiment result through the classical domain and neighborhood domain extensionto narrow down and find the optimum capability planning. The advantages of the proposed ETMinclude: (1) the optimization process is subject to the specifications of practical equipment; (2) fastcomputing speed; and (3) the results could be directly applied on the real situation.

2. System Model

The stand-alone power system is divided into electricity generation, energy storage and standbyenergy. Electricity generation is formed by the wind power and PV power system using the kineticenergy of wind and the solar luminous energy. Although the wind and solar complementarysystems make up for each other’s deficiencies as shown in Figure 1, where the natural energygenerating capacity cannot create a continuous power supply all the time due to weather variations.Therefore, energy storage equipment is required. When the electricity generated is enough, it shalllaunch the electrolyzer to produce the hydrogen, and store the latter in the hydrogen tank. When thegenerating capacity is insufficient, it could use the hydrogen in the hydrogen tank and the fuel cell tosatisfy the load as shown in Figure 2. The dump load is also included as a component in the systemto consume the excess electricity when renewable power is abundant and the storage system is fullycharged. In order to make a system device with the most cost-saving and most efficient power supplycapacity, the system capability configuration must operate optimally.

Energies 2016, 9, 174 2 of 17

for the size optimization of a PV/wind hybrid energy conversion system with battery storage. The

drawback of this method is that it takes too much computing time and the computing results needs

to be adjusted through experience to fit in the specifications of the common practical products.

Yang et al. [9] applied genetic algorithm to calculate the optimal system configuration with

relative computational simplicity. Some commercial software like HOMER [10], IHOGA [11] were

developed to figure out the optimal allocation. Based on the simulated weather data, this study

estimates the possibility of loss of load probability (LOLP) for one or more years. Taguchi quality

engineering was developed and advocated by Dr. Taguchi Genichi [12] in 1950. The orthogonal array

experiment design and analysis of variance (ANOVA) were used for the analysis with a limited

experimental data to effectively improve product quality [12]. The Taguchi method is widely applied

in the manufacturing field, but this study is the first to apply the Taguchi method in the optimization

of hybrid power systems.

This paper combines the extension theory with the Taguchi method, selects the proper design

factors and levels via the orthogonal array to greatly decrease the number of experiments to acquire

the optimum experiment result through the classical domain and neighborhood domain extension to

narrow down and find the optimum capability planning. The advantages of the proposed ETM

include: (1) the optimization process is subject to the specifications of practical equipment; (2) fast

computing speed; and (3) the results could be directly applied on the real situation.

2. System Model

The stand‐alone power system is divided into electricity generation, energy storage and standby

energy. Electricity generation is formed by the wind power and PV power system using the kinetic

energy of wind and the solar luminous energy. Although the wind and solar complementary systems

make up for each other’s deficiencies as shown in Figure 1, where the natural energy generating

capacity cannot create a continuous power supply all the time due to weather variations. Therefore,

energy storage equipment is required. When the electricity generated is enough, it shall launch the

electrolyzer to produce the hydrogen, and store the latter in the hydrogen tank. When the generating

capacity is insufficient, it could use the hydrogen in the hydrogen tank and the fuel cell to satisfy the

load as shown in Figure 2. The dump load is also included as a component in the system to consume

the excess electricity when renewable power is abundant and the storage system is fully charged. In

order to make a system device with the most cost‐saving and most efficient power supply capacity,

the system capability configuration must operate optimally.

Figure 1. System architecture. Figure 1. System architecture.

Energies 2016, 9, 174 3 of 17

2.1. The Objective Function

Normally, this system uses wind power and PV power system, and makes use of wind-lightcomplementarity feature to generate electricity. If wind turbines and solar cell cannot supply ratedload, then determine whether there is hydrogen can be used, and then use hydrogen to satisfy load.The loss of load hours βH will increase if hydrogen is not sufficient. The calculation process is shownin Figure 2, where PW is power capacity of wind power, PPV is power capacity of the PV power system,PL is load capacity.

Energies 2016, 9, 174 3 of 17

2.1. The Objective Function

Normally, this system uses wind power and PV power system, and makes use of wind‐light

complementarity feature to generate electricity. If wind turbines and solar cell cannot supply rated

load, then determine whether there is hydrogen can be used, and then use hydrogen to satisfy load.

The loss of load hours βH will increase if hydrogen is not sufficient. The calculation process is shown

in Figure 2, where PW is power capacity of wind power, PPV is power capacity of the PV power system,

PL is load capacity.

Figure 2. Operation flowchart of the proposed stand‐alone power system.

The above βH of the system can be used to calculate the loss of load probability (LOLP) βP.

Bringing βP into the algorithm is to solve the contradiction of load capacity, loss of load, and cost.

The calculation of βP is shown in Equation (1):

HP T

(1)

βP per hour is often expressed as 0 or 1. When βP is 1 in one hour, it indicates that the power

system cannot supply the rated load [3]. The more times a 1 is recorded in total time T, the poorer the

reliability of the independent power system.

Many methods and cost functions are available, and differences exist among them. Considering

a stand‐alone power system, especially in a remote area that the utility power grid cannot reach, the

possibility of LOLP and the return on investment are the most two important factors in Taiwan.

According to the cost planning of available products, cost Bc and LOLP are equally important. The

sum of weights is 1, and the weights of cost ( 1 ) and LOLP ( 2 ) are set as 0.5. The weights vary

Figure 2. Operation flowchart of the proposed stand-alone power system.

The above βH of the system can be used to calculate the loss of load probability (LOLP) βP.Bringing βP into the algorithm is to solve the contradiction of load capacity, loss of load, and cost.The calculation of βP is shown in Equation (1):

βP “βHT

(1)

βP per hour is often expressed as 0 or 1. When βP is 1 in one hour, it indicates that the powersystem cannot supply the rated load [3]. The more times a 1 is recorded in total time T, the poorer thereliability of the independent power system.

Many methods and cost functions are available, and differences exist among them. Consideringa stand-alone power system, especially in a remote area that the utility power grid cannot reach,the possibility of LOLP and the return on investment are the most two important factors in Taiwan.According to the cost planning of available products, cost Bc and LOLP are equally important. The sum

Energies 2016, 9, 174 4 of 17

of weights is 1, and the weights of cost (λ1) and LOLP (λ2) are set as 0.5. The weights vary dependingon different system requirements. The objective function of the paper is shown in Equation (2):

MinF “ÿ

βCppuq ˆ λ1 `βpppuq ˆ λ2 (2)

where:

Bcppuq “Bcpkq ´ Bc min

Bc max ´ Bc min(3)

Bpppuq “Bppkq ´ Bp min

Bp max ´ Bp min(4)

In addition, we also calculate the investment recovery period of the total power generation, andthe total recovery period is shown in Equation (5):

YC “Ct

pEp ˆ Gp ´ Ct ˆ Cpq(5)

where YC is the return period on investment, Ct is the setup cost, Ep is total power generation, Gp isthe power unit cost, and Cp is bank interest rate in Taiwan.

2.2. Mathematical Models of Solar Photovoltaic

Silicon is the most easily obtainable material used for manufacturing semiconductors, and it isplentiful on the planet. Currently silicon is the main material for solar cells. Photovoltaic modulesare major formed by serially/parallelly connecting solar cells together, and usually photovoltaicmodule structures contain large PN junction diodes. When solar cells are illuminated by sunlight,they will convert light energy into electrical energy. Being serially or parallelly connected in a module,a solar cell will have a different impact on the current and voltage of array output. When a parallellyconnected solar cell is added, it will increase the array current; when a serially connected solar cellis added, it will increase the array voltage, where Iph is the current generated by solar cell when itis illuminated by sunlight, [4], and the equations of the solar energy equivalent circuit are shown inEquations (6) and (7):

IPH “ Zˆ ISCp1` Ko ˆ pTPV ´ Trqq (6)

PPV “ np IPHV ´ np IsatV„

epq

BKTPV

Vnsq ´ 1

(7)

where Z is illuminance, ISC is a current on a light intensity 1000 W/m2 and ambient temperature298 K, Isat is the reverse saturation current, np is the number of solar cells connected in parallel, ns isthe number of solar cells in series, Ko is the short circuit current correction coefficient, B is an idealparameter which is 1, K is the Boltzmann constant (1.3806 ˆ 10´23 J/K), Tpv and Tr are the real andreference temperature of the solar cell.

2.3. Mathematical Models of Wind Power

Wind turbines obtain kinetic energy from wind, mainly from blades touched by air flow whenthe Earth rotates. Wind energy will rotate blades; the wind turbine blade shaft will convert windenergy into useful mechanical energy, and this is followed by the conversion of mechanical energyinto electrical energy. Due to the aerodynamic effect, the blade rotor will be rotated by the force ofthe wind, and the aerodynamic force is divided into lift force and drag force to produce torque onthe blades [5]. The output power of the wind turbine is converted into mechanical energy, and theequation generated by wind power generation is shown as in Equation (8) [6]:

PA “12ρπR2V3

w (8)

Energies 2016, 9, 174 5 of 17

When the wind speed is constant, the wind turbine power output is proportional to the square ofwind turbine’s diameter. When a wind turbine’s diameter is fixed, its power output is proportional tothe cube of wind speed, when wind diameter and wind speed remain unchanged. A wind turbine’spower output is proportional to CP [13], and the wind output equation is shown as in Equation (9):

Pw “ Cp ˆ pTSRq ˆ PA (9)

where ρ the is air density (kg/m3), R is the radius of the rotor blade (m), VW is wind speed (m/s), TSRis the wind turbine’s blade tip speed ratio, and Cp is the wind turbine’s performance index.

When the wind speed reaches Vo f f , the turbine will generally lock the rotor and suspend powergeneration in order to prevent the wind turbine from operating at speeds which may damage theblades, and this wind speed is called the shutdown wind speed. On the contrary, when the wind speedis lower than Von, although the wind turbine can continue generating electricity, usually its operationwill be suspended to prevent the wind turbine from disconnecting from/connecting to the powersystem too frequently, and this wind speed is called the cut-in wind speed. Based on Equation (10), thetypical curve of the wind power system is shown in Figure 3 [14]:

PwpVWq “

$

’

’

&

’

’

%

0

Cp ˆ pTSRq ˆ p12ρπR2V3

wq

0

ˇ

ˇ

ˇ

ˇ

ˇ

ˇ

ˇ

Vw ă Von

Von ď Vw ď Vo f fVw ą Vo f f

,

/

/

.

/

/

-

(10)

where Vw is the current wind speed, Von is the starting wind speed, and Vo f f is the ending wind speed.

Energies 2016, 9, 174 5 of 17

2 31

2A wP R V (8)

When the wind speed is constant, the wind turbine power output is proportional to the square

of wind turbine’s diameter. When a wind turbine’s diameter is fixed, its power output is proportional

to the cube of wind speed, when wind diameter and wind speed remain unchanged. A wind turbine’s

power output is proportional to CP [13], and the wind output equation is shown as in Equation (9):

( )w p SR AP C T P (9)

where the is air density (kg/m3), R is the radius of the rotor blade (m), VW is wind speed (m/s), TSR

is the wind turbine’s blade tip speed ratio, and Cp is the wind turbine’s performance index.

When the wind speed reaches Voff, the turbine will generally lock the rotor and suspend power

generation in order to prevent the wind turbine from operating at speeds which may damage the

blades, and this wind speed is called the shutdown wind speed. On the contrary, when the wind

speed is lower than Von, although the wind turbine can continue generating electricity, usually its

operation will be suspended to prevent the wind turbine from disconnecting from/connecting to the

power system too frequently, and this wind speed is called the cut‐in wind speed. Based on

Equation (10), the typical curve of the wind power system is shown in Figure 3 [14]:

2 3

0

1( ) ( ) ( )

20

w on

w W p SR w on w off

w off

V V

P V C T R V V V V

V V

(10)

where Vw is the current wind speed, Von is the starting wind speed, and Voff is the ending wind speed.

Figure 3. Typical curve of a wind power system [14].

2.4. Mathematical Models of Fuel Cell

Fuel cells are a kind of device that converts chemical energy into electrical energy. The voltage

of a single fuel cell is too low to be practical (approximately 0.6 to 0.8 volts), so it is necessary to

serially connect fuel cells together in order to raise the power output voltage [15]. Fuel cells are subject

to some losses when converting chemical energy into electrical energy. For fuel cells, the power

output can reach its highest value only in a reversible state, but the power output is less than the ideal

power because it is subject to some restrictions in actual operation. The fuel cells used in this study

are proton exchange membrane fuel cells, and the expression governing these fuel cells is given by

Equation (11):

FC Nermo act ohmic conV =E V V V (11)

Figure 3. Typical curve of a wind power system [14].

2.4. Mathematical Models of Fuel Cell

Fuel cells are a kind of device that converts chemical energy into electrical energy. The voltage ofa single fuel cell is too low to be practical (approximately 0.6 to 0.8 volts), so it is necessary to seriallyconnect fuel cells together in order to raise the power output voltage [15]. Fuel cells are subject to somelosses when converting chemical energy into electrical energy. For fuel cells, the power output canreach its highest value only in a reversible state, but the power output is less than the ideal powerbecause it is subject to some restrictions in actual operation. The fuel cells used in this study are protonexchange membrane fuel cells, and the expression governing these fuel cells is given by Equation (11):

VFC “ ENermo ´Vact ´Vohmic ´Vcon (11)

Energies 2016, 9, 174 6 of 17

where, VFC is output voltage, ENermo is the voltage of the reversible cell (thermal electric potential offuel cell), Vact is the activation overpotential, Vohmic is the ohmic overpotential (ohmic overvoltageproduced by the resistance effect of electrons passing through the electrode and protons passingthrough the exchange membrane), Vcon is the concentration overpotential (concentration overpotentialcaused by the reduction of reaction gas concentration).

In Equation (11), ENermo is the thermodynamic reversible voltage output by a single open circuit.Results obtained through the Nernst equation at a reference temperature of 25 ˝C are shown inEquation (12):

ENernst “∆G2F

`∆S2FpT´ T0q `

RT2F

„

lnpPH2q `12

lnpPO2q

(12)

where, the Gibbs Free Energy change value is indicated by ∆G (J/mol), F is the Faraday constant(96487C), ∆S is the entropy value change, R is the universal gas constant (8.314 J/K¨mol), PH2 and PO2

are the pressures of hydrogen and oxygen needed by fuel cell, T is the working temperature of fuelcell and T0 is the reference temperature. ∆G, ∆S and T0 under standard temperature and standardatmospheric pressure are substituted into Equation (12) to obtain the simplified Equation (13):

ENermo “ 1.229´ 0.85ˆ 10´3pT´ 298.15q ` 4.31ˆ 10´5 Tˆ rlnpPH2q ` 0.5lnpPO2qs (13)

Vact in Equation (11) is the voltage drop produced by the activated anode and cathode, which isshown in Equation (14):

Vact “ rξ1 ` ξ2T` ξ3 lnpCO2q ` ξ4T lnpIFCqs (14)

IFC in Equation (14) is the current of the fuel cell, CO2 (atm) the oxygen concentration and ξi

(i = 1–4) are the characteristic coefficients of all fuel cells. Vohmic voltage drop in Equation (11) willcause a voltage loss of the fuel cell due to the resistance in charge transfer, in conformity with Ohm’sLaw, see Equation (15):

Vohmic “ IFCpRM ` Rcq (15)

The Ohmic loss of a fuel cell can be minimized through use of a thin electrolyte membrane andconductive materials. RC in Equation (15) is the resistance of the contact electron current, which isdeemed as constant in the literature since it can’t be obtained within the working temperature ofproton exchange membrane fuel cell. RM is the resistance of the proton exchange membrane, whichcan be obtained through an empirical formula, see Equation (16):

RM “ρM ˆ l

Ac(16)

ρM “ ab

a “ 181.6

«

1` 0.03IFCAC

` 0.062ˆ

T303

˙2 ˆ IFCAC

˙2.5ff

b “„

ψ´ 0.634´ 3ˆ

IFCAC

˙

exp„

4.18ˆ

T´303T

˙

(17)

AC in Equation (16) is the activation area of the fuel cell (cm2), while the thickness of the electrolytemembrane is expressed by l (cm), and the resistance of the electrolyte membrane ρM (Ω¨cm) indicatesthe coefficient, which is given by Equation (17). Parameter Ψ is the water content of membrane.The value of Ψ falling at 14 indicates an ideal relative humidity of 100%. However, the value of Ψgenerally falls in the oversaturation region of 22–23, which is an adjustable parameter.

The voltage loss produced by a fuel cell in large-current operation is Vcon, indicating theconcentration loss caused by diffusion limited mass transfer, see Equation (18):

Vcon “ Bˆ lnˆ

1´J

Jmax

˙

(18)

Energies 2016, 9, 174 7 of 17

In Equation (18), B is the constant of a fuel cell of any type, which is determined by the fuel celland its operation status. J is the current density (mA/cm2) produced by the fuel cell, Jmax is the largestcurrent density, which is determined between 500 and 1500 (mA/cm2). Due to electrochemical action,the nitric oxide and carbonic oxide discharged by fuel cell is quite abundant. Thus, the inventionof hydrogen energy converters with economic effectiveness and high efficiency will promote futurecommercial use. A typical 200 W fuel cell characteristic curve is as shown in Figure 4.

Energies 2016, 9, 174 7 of 17

max

ln 1con

JV B

J

(18)

In Equation (18), B is the constant of a fuel cell of any type, which is determined by the fuel cell

and its operation status. J is the current density (mA/cm2) produced by the fuel cell, Jmax is the largest

current density, which is determined between 500 and 1500 (mA/cm2). Due to electrochemical action,

the nitric oxide and carbonic oxide discharged by fuel cell is quite abundant. Thus, the invention of

hydrogen energy converters with economic effectiveness and high efficiency will promote future

commercial use. A typical 200 W fuel cell characteristic curve is as shown in Figure 4.

(a)

(b)

(c)

Figure 4. Characteristic curves of the tested fuel cell. (a) Characteristic curves of power‐current under

different humidity; (b) Characteristic curves of power‐current under different hydrogen pressures;

(c) Characteristic curves of power‐current under different temperatures.

2.5. Objective Function Setting and Limitation

A well‐optimized system will reduce costs to a minimum while satisfying the load demand. The

targeted function of this paper is as Equation (2) [16], in which, the commercial market product has

the consumption costs shown in Table 1.

Figure 4. Characteristic curves of the tested fuel cell. (a) Characteristic curves of power-current underdifferent humidity; (b) Characteristic curves of power-current under different hydrogen pressures;(c) Characteristic curves of power-current under different temperatures.

2.5. Objective Function Setting and Limitation

A well-optimized system will reduce costs to a minimum while satisfying the load demand.The targeted function of this paper is as Equation (2) [16], in which, the commercial market producthas the consumption costs shown in Table 1.

Energies 2016, 9, 174 8 of 17

Table 1. Cost of Each Design Factor.

Equipment Name Equipment Cost Unit Capacity Cost (USD)

Wnd power CW 1000 W 688PV power system CPV 230 W 565

Hydrogen tank CHT 40 L 235Fuel Cell CFC 100 W 1250

Electrolyzer CE 100 W 1563

According to the regular market specifications of power equipment, the minimum power supplyneeds to satisfy the Equation (19) as follows:

ÿ

i“1

pPW ` PV ` PFCq ě PLOAD ` PE ` Pdm (19)

3. Extension Taguchi Method

To optimize the allocation of equipment capacity, this paper proposes an optimization methodcombining extension theory and the Taguchi method. In 1983, the concept of extension theorywas proposed to solve the contradictions and incompatibility problems. Based on some propertransformation, some problems that cannot be directly solved by given conditions may becomeeasier or solvable [17]. One of the commonly used techniques in engineering fields is the Laplacetransformation. The concept of fuzzy sets is a generalization of well-known standard sets to extendtheir application fields. Therefore, the concept of an extension set is to extend the fuzzy logic valuefrom [0,1] to (´8, 8), which allows us to define any data in the domain and has given promisingresults in many fields [18–20]. Extension theory is used to calculate the comprehensive correlativedegree of each group of the simulation experiment, then the analytical method average value of theTaguchi method is used to select the optimum capability configuration system and assess whetherthe selected group is the optimum result [21]. If the output result is optimal, then the process is done;otherwise, the extension correlative degree shall be recalculated. The advantage of this method is theability to find the optimum solution in the shortest time and to limit the number of experiment trialsneeded. The calculation steps are as described below:

Step 1. Setting the level and design

This paper selects five three-level design factors in order to conform to the basic principle ofan L18 orthogonal array. The design factors are used to reduce the power shortage probability whilenot ignoring the equipment cost. Thus the price is also set as a design objective. The design factorand the corresponding levels are shown in Table 2. The five three-level design factors are wind power,PV power, hydrogen tank, fuel cell, and electrolyzer. For example, the numbers of the wind powerequipment designed for three corresponding levels are 1, 2, and 3. The numbers of equipment of eachlevel for the design factors in Table 2 are subject to adjustment according to the setup costs, and theoptimal results could be different.

Table 2. The design factors and their corresponding levels.

Design Factors Level 1 Level 2 Level 3

Wind power QW 1 2 3PV power QPV 3 4 5

Hydrogen tank QHT 10 15 20Fuel Cell QFC 3 4 5

Electrolyzer QE 2 3 4

Energies 2016, 9, 174 9 of 17

Step 2. Using the proper orthogonal array

The Taguchi method significantly reduces the number of experimental configurations to bestudied. Originally, there are 35 = 243 experiments in the full factorial design, and the L18 orthogonalarray can accommodate five three-level design factors in only 18 experiments to efficiently collectexperimental results to find the optimized factor combination. The proposed L18 orthogonal arraytable is shown in Table 3.

Table 3. The orthogonal array of L18.

L18 QW QPV QHT QFC QE

1 1 1 1 1 12 1 2 2 2 23 1 3 3 3 34 2 1 1 2 25 2 2 2 3 36 2 3 3 1 17 3 1 2 1 38 3 2 3 2 19 3 3 1 3 210 1 1 3 3 211 1 2 1 1 312 1 3 2 2 113 2 1 2 3 114 2 2 3 1 215 2 3 1 2 316 3 1 3 2 317 3 2 1 3 118 3 3 2 1 2

Step 3. Calculating the initial result

According to the experimental combinations in the orthogonal array, Matlab is applied to generatethe 18 different electric power generation system quantities in the stand-alone power system whichcould simulate 18 different systems, and calculate each electric power generation system quantity priceand power shortage hours, respectively. The results are shown in Table 3.

Step 4. Calculating the per-unit values

The simulated experiment data cannot be compared or calculated for the unit reference valuedifference, thus it requires per-unit value calculation to get the physical quantity relative to thereference value [22]. The necessary equation is as shown in Equations (3) and (4).

Step 5. Establishing the matter-element

In extension theory, N is given as the name of an object for the definition of the object, and thequantity value of its character c is v. The orderly combination is taken as the fundamental elementdescribing the object, namely the matter-element. The name of the object may not be expressed by thesingle matter-element, but expressed by the multi-dimensional matter-element. The multi-dimensionalmatter-element assumes the object has n characters respectively c1, c2 . . . cn, and its correspondingquantity value is v1, v2 . . . vn. Putting the above Sg

pupkq into v, the multi-dimensional matter-elementexpression is as shown in Equation (20), and the second group is placed after calculating the correlationfunctional value:

Rt “

«

N βcppuq Vt1βpppuq Vt2

ff

(20)

Energies 2016, 9, 174 10 of 17

Step 6. Establishing the classical domain and neighborhood domain

In extension theory, the positional relation of one point to two sections must be considered.If Vo “ xa, by and Vp “ xc, dy are two sections in the real domain, it shall firstly confirm the classicaldomain and the neighborhood domain before getting the matter-element in. Establishing the system’smatter-element model, the defined classical domain is as Equation (21) and the neighborhood domainis as Equation (22). Defining the range according to the design factor demand, in extension theory, theclassical domain is the expected value, and the neighborhood domain is the overall range.

Ro “

«

systemβpppuq x0.7713, 1.1739yβcppuq x0.5181, 1.1242y

ff

(21)

Rp “

«

systemβpppuq x0.1539, 1.1739yβcppuq x0.1827, 1.1242y

ff

(22)

Step 7. Calculating the correlation values

According to the definition of distance, we confirm the correlation function and use Equation (23)for assessment. If the value to be measured is within the classical domain, it shall be calculated bythe method of Equation (23a); if the value to be measured is outside the classical domain, it shall becalculated in the method of Equation (23b) to get the correlation function.

Kipviq “

$

’

’

’

&

’

’

’

%

´ρpvi, Vojqˇ

ˇVojˇ

ˇ

, vi P Voj ¨ ¨ ¨ ¨ ¨ ¨ paq

ρpvi, Vojq

ρpvi, Vpjq ´ ρpvi, Vojq, vi R Voj ¨ ¨ ¨ ¨ ¨ ¨ pbq

(23)

The proposed extension relation function can be shown as Figure 5, where 0 ď K(v) ď 1corresponds to the normal fuzzy set. It describes the degree to which v belong to V. When K(v) < 0, itindicates the degree to which v does not belong to V.

Energies 2016, 9, 174 10 of 17

In extension theory, the positional relation of one point to two sections must be considered. If

,oV a b and ,pV c d are two sections in the real domain, it shall firstly confirm the classical

domain and the neighborhood domain before getting the matter‐element in. Establishing the system’s

matter‐element model, the defined classical domain is as Equation (21) and the neighborhood domain

is as Equation (22). Defining the range according to the design factor demand, in extension theory,

the classical domain is the expected value, and the neighborhood domain is the overall range.

( ) 0.7713,1.1739

( ) 0.5181,1.1242p

oc

puR system

pu

(21)

( ) 0.1539,1.1739

( ) 0.1827,1.1242p

pc

puR system

pu

(22)

Step 7. Calculating the correlation values

According to the definition of distance, we confirm the correlation function and use Equation (23)

for assessment. If the value to be measured is within the classical domain, it shall be calculated by the

method of Equation (23a); if the value to be measured is outside the classical domain, it shall be

calculated in the method of Equation (23b) to get the correlation function.

( , ), ( )

( )( , )

, ( )( , ) ( , )

i oji oj

oj

i i

i oji oj

i pj i oj

v Vv V a

VK v

v Vv V b

v V v V

(23)

The proposed extension relation function can be shown as Figure 5, where 0 ( ) 1K v

corresponds to the normal fuzzy set. It describes the degree to which v belong to V. When ( ) 0K v ,

it indicates the degree to which v does not belong to V.

Figure 5. The proposed extension relation function.

Step 8. Comprehensive correlation degree

In the stand‐alone power system, normally, the power shortage hours have higher priority than

cost and the correlation degree function 1 is set at 0.62 and 2 is at 0.38 as in Equation (24), the

result is given as shown in Table 4.

1

aij bij

apj

K(v)

bpjv

-1

Figure 5. The proposed extension relation function.

Step 8. Comprehensive correlation degree

In the stand-alone power system, normally, the power shortage hours have higher priority thancost and the correlation degree function α1 is set at 0.62 and α2 is at 0.38 as in Equation (24), the resultis given as shown in Table 4.

Kppq “nÿ

i“1

αjKipviq (24)

Energies 2016, 9, 174 11 of 17

Table 4. This research’s level and design factor table.

L18 βP Cost (USD)

1 0.35927 11,6092 0.32500 16,1623 0.27777 20,7154 0.21527 15,1105 0.15833 19,6636 0.14305 15,7777 0.10416 17,2868 0.07316 17,1509 0.08333 18,17410 0.36388 18,02211 0.33472 15,30012 0.28055 15,16413 0.19444 15,97214 0.15972 16,77515 0.18472 17,80316 0.09444 19,71117 0.09305 16,05018 0.07638 16,853

Step 9. Average value analysis

Average value analysis uses the Taguchi method to calculate the influence of the factor effectsto confirm the optimal factor combination by taking the average value analysis of levels for eachfactor by normalizing each level of comprehensive correlative degree of Table 5 according to the L18

orthogonal array grade in Table 4 to calculate the average value. The average value analysis is asshown in Equation (25), and the result is as shown in Table 6.

m “1n

nÿ

i“1

Kppq (25)

Table 5. Comprehensive Correlative Degree.

L18 Comprehensive Correlative Degree

1 ´0.026972 ´0.150523 ´0.743384 0.1104585 ´0.396596 0.1625567 0.0596348 0.0682949 ´0.07669

10 ´0.4539311 ´0.4539312 0.00663413 0.06450514 0.04647815 ´0.1481616 ´0.3032217 0.21692618 0.118986

Energies 2016, 9, 174 12 of 17

Step 10. Solution of optimum system parameter

Through the analysis of factor effects, the maximum stand-alone power system optimumallocation can be selected through average value analysis of the Taguchi method as shown in Table 6.

Table 6. Optimum Parameters of the ETM.

Parameter Load QW QPV QHT QFC QE

Level 2 2 1 1 1 2Optimum quantity 500 W 2 3 10 3 3

4. Simulated Result and Discussion

This study’s simulation adopts the weather data of the Wuci area in central Taiwan in 2014,provided by Taiwan Central Weather Bureau, to compare the wind speed and the illumination ofthe winter and summer, as shown in Figures 6 and 7. The sun radiates directly on the SouthernHemisphere, and the illumination is small and the wind power is strong in Taiwan. The load power(PL), wind (pw) power and PV power (PPV) of the calculated optimum allocation are shown inFigure 8. If the amount is less than the generating capacity load, fuel cells will supply the load power,as shown in Figure 9. In summer, the load power (PL), wind (pw) power and PV power (PPV) of thecalculated optimum allocation are also shown in Figure 10. If the amount is less than the generatingcapacity load, fuel cells power will be used to supply the load, as shown in Figure 11. In summer,the sun directly radiates on the Northern Hemisphere, the PV power increases in energy. However,there are usually windless situations in the summer, and the operation of the wind power generatoris suspended during typhoon season, thus wind power generation shall decrease. Generally, duringthe daytime, solar energy is abundant and the use of the fuel cells is not required, and the system isable to produce hydrogen for extra needs. At night, there is no illumination, and only wind powergeneration provides energy. The fuel cells shall use the hydrogen to support the load. At night, inaddition the home electricity usage increases, therefore, in the stand-alone power system, the fuel celldoes not need much generating capacity, but needs more hydrogen tanks for standby to supplementthe load demand.

Energies 2016, 9, 174 12 of 17

11 −0.45393

12 0.006634

13 0.064505

14 0.046478

15 −0.14816

16 −0.30322

17 0.216926

18 0.118986

Table 6. Optimum Parameters of the ETM.

Parameter Load QW QPV QHT QFC QE

Level 2 2 1 1 1 2

Optimum quantity 500 W 2 3 10 3 3

4. Simulated Result and Discussion

This study’s simulation adopts the weather data of the Wuci area in central Taiwan in 2014,

provided by Taiwan Central Weather Bureau, to compare the wind speed and the illumination of the

winter and summer, as shown in Figures 6 and 7. The sun radiates directly on the Southern

Hemisphere, and the illumination is small and the wind power is strong in Taiwan. The load power (PL),

wind (pw) power and PV power (PPV) of the calculated optimum allocation are shown in Figure 8. If the

amount is less than the generating capacity load, fuel cells will supply the load power, as shown in

Figure 9. In summer, the load power (PL), wind (pw) power and PV power (PPV) of the calculated

optimum allocation are also shown in Figure 10. If the amount is less than the generating capacity

load, fuel cells power will be used to supply the load, as shown in Figure 11. In summer, the sun

directly radiates on the Northern Hemisphere, the PV power increases in energy. However, there are

usually windless situations in the summer, and the operation of the wind power generator is

suspended during typhoon season, thus wind power generation shall decrease. Generally, during

the daytime, solar energy is abundant and the use of the fuel cells is not required, and the system is

able to produce hydrogen for extra needs. At night, there is no illumination, and only wind power

generation provides energy. The fuel cells shall use the hydrogen to support the load. At night, in

addition the home electricity usage increases, therefore, in the stand‐alone power system, the fuel cell

does not need much generating capacity, but needs more hydrogen tanks for standby to supplement

the load demand.

(a)

0 20 40 60 80 100 120 140 1600

2

4

6

8

10

12

14

16

Figure 6. Cont.

Energies 2016, 9, 174 13 of 17Energies 2016, 9, 174 13 of 17

(b)

Figure 6. The curves of the typical weekly wind speed (a) winter; (b) summer.

(a)

(b)

Figure 7. The curves of the typical solar illuminance (a) winter; (b) summer.

0 20 40 60 80 100 120 140 1600

1

2

3

4

5

6

7

8

Time (hr)0 20 40 60 80 100 120 140 160

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 20 40 60 80 100 120 140 1600

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Figure 6. The curves of the typical weekly wind speed (a) winter; (b) summer.

Energies 2016, 9, 174 13 of 17

(b)

Figure 6. The curves of the typical weekly wind speed (a) winter; (b) summer.

(a)

(b)

Figure 7. The curves of the typical solar illuminance (a) winter; (b) summer.

0 20 40 60 80 100 120 140 1600

1

2

3

4

5

6

7

8

Time (hr)0 20 40 60 80 100 120 140 160

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 20 40 60 80 100 120 140 1600

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Figure 7. The curves of the typical solar illuminance (a) winter; (b) summer.

Energies 2016, 9, 174 14 of 17Energies 2016, 9, 174 14 of 17

Figure 8. The power curves of the power system in the first week of December.

Figure 9. The curves of the hydrogen capability and fuel cell power in the first week of December.

Figure 10. The power curves of the power system in the first week of June.

0 20 40 60 80 100 120 140 1600

500

1000

1500

2000

2500

3000

PLPPVPW

0 20 40 60 80 100 120 140 1600

100

200

300

400

500

600

700

800

H2PFC

0 20 40 60 80 100 120 140 1600

200

400

600

800

1000

1200

PLPPVPW

Figure 8. The power curves of the power system in the first week of December.

Energies 2016, 9, 174 14 of 17

Figure 8. The power curves of the power system in the first week of December.

Figure 9. The curves of the hydrogen capability and fuel cell power in the first week of December.

Figure 10. The power curves of the power system in the first week of June.

0 20 40 60 80 100 120 140 1600

500

1000

1500

2000

2500

3000

PLPPVPW

0 20 40 60 80 100 120 140 1600

100

200

300

400

500

600

700

800

H2PFC

0 20 40 60 80 100 120 140 1600

200

400

600

800

1000

1200

PLPPVPW

Figure 9. The curves of the hydrogen capability and fuel cell power in the first week of December.

Energies 2016, 9, 174 14 of 17

Figure 8. The power curves of the power system in the first week of December.

Figure 9. The curves of the hydrogen capability and fuel cell power in the first week of December.

Figure 10. The power curves of the power system in the first week of June.

0 20 40 60 80 100 120 140 1600

500

1000

1500

2000

2500

3000

PLPPVPW

0 20 40 60 80 100 120 140 1600

100

200

300

400

500

600

700

800

H2PFC

0 20 40 60 80 100 120 140 1600

200

400

600

800

1000

1200

PLPPVPW

Figure 10. The power curves of the power system in the first week of June.

Energies 2016, 9, 174 15 of 17Energies 2016, 9, 174 15 of 17

Figure 11. The curves of the hydrogen capability and fuel cell power in the first week of June.

The optimum allocation into the weekly and monthly average data shall be simulated to

compare the probability of shortage allocated for the four types of algorithms, as shown in Figure 12.

Figure 12. Annual Power Shortage Probability (LOLP).

The extension hierarchical method firstly adopts the extension classical domain and

neighborhood domain, taking the matter‐element to be measured to calculate the correlation function

and extension distance, and forming the contrast matrix by the empirical law to establish hierarchical

analysis method to conform to the requirement of consistency and gain weight. Comparing AHP and

the Extension Taguchi Method (ETM), although the power shortage hours of AHP are good, the

allocation of the ETM is better than with other optimum methods.

5. Conclusions

This study aims to compare the performance of the extension theory, extension AHP theory,

ETM and Analytic Hierarchy Process (AHP) in determining the optimized capacity allocation of

stand‐alone power systems based on calculating the power shortage probability and construction

cost. Through the experiments setting of orthogonal array, ETM is used to significantly reduce the

number of experiments to efficiently verify the best factor combination for the capacity allocation.

Results show that the ETM’s allocation parameters are the best among all models, and the conclusion

of this paper are as follows:

0 20 40 60 80 100 120 140 1600

100

200

300

400

500

600

700

800

H2PFC

Figure 11. The curves of the hydrogen capability and fuel cell power in the first week of June.

The optimum allocation into the weekly and monthly average data shall be simulated to comparethe probability of shortage allocated for the four types of algorithms, as shown in Figure 12.

Energies 2016, 9, 174 15 of 17

Figure 11. The curves of the hydrogen capability and fuel cell power in the first week of June.

The optimum allocation into the weekly and monthly average data shall be simulated to

compare the probability of shortage allocated for the four types of algorithms, as shown in Figure 12.

Figure 12. Annual Power Shortage Probability (LOLP).

The extension hierarchical method firstly adopts the extension classical domain and

neighborhood domain, taking the matter‐element to be measured to calculate the correlation function

and extension distance, and forming the contrast matrix by the empirical law to establish hierarchical

analysis method to conform to the requirement of consistency and gain weight. Comparing AHP and

the Extension Taguchi Method (ETM), although the power shortage hours of AHP are good, the

allocation of the ETM is better than with other optimum methods.

5. Conclusions

This study aims to compare the performance of the extension theory, extension AHP theory,

ETM and Analytic Hierarchy Process (AHP) in determining the optimized capacity allocation of

stand‐alone power systems based on calculating the power shortage probability and construction

cost. Through the experiments setting of orthogonal array, ETM is used to significantly reduce the

number of experiments to efficiently verify the best factor combination for the capacity allocation.

Results show that the ETM’s allocation parameters are the best among all models, and the conclusion

of this paper are as follows:

0 20 40 60 80 100 120 140 1600

100

200

300

400

500

600

700

800

H2PFC

Figure 12. Annual Power Shortage Probability (LOLP).

The extension hierarchical method firstly adopts the extension classical domain and neighborhooddomain, taking the matter-element to be measured to calculate the correlation function and extensiondistance, and forming the contrast matrix by the empirical law to establish hierarchical analysis methodto conform to the requirement of consistency and gain weight. Comparing AHP and the ExtensionTaguchi Method (ETM), although the power shortage hours of AHP are good, the allocation of theETM is better than with other optimum methods.

5. Conclusions

This study aims to compare the performance of the extension theory, extension AHP theory, ETMand Analytic Hierarchy Process (AHP) in determining the optimized capacity allocation of stand-alonepower systems based on calculating the power shortage probability and construction cost. Through theexperiments setting of orthogonal array, ETM is used to significantly reduce the number of experimentsto efficiently verify the best factor combination for the capacity allocation. Results show that the ETM’sallocation parameters are the best among all models, and the conclusion of this paper are as follows:

Energies 2016, 9, 174 16 of 17

(1) This paper proposes an optimum capability planning combining the extension theories andTaguchi method to solve the SAPS optimum problem; the proposed model reaches the optimalsolution efficiently.

(2) The proposed ETM needs to firstly specify the level and design factor. To guarantee theexperimental results, the domain knowledge and expert experience are required when setting thedesign factors and their levels.

(3) Since the specifications of commercial equipment are pre-specified, the optimum system allocationcan be practically achieved by using the proposed ETM.

(4) With fast computation speed, the proposed ETM can be easily applied to mathematical modelswith complicated objective functions and power systems, and the generated results could bedirectly implemented on practical equipment without further experimental adjustments.

Acknowledgments: This work was supported in part by the Ministry of Science and Technology, Taiwan.Under the grant no: 104-2623-E-167-001-ET.

Author Contributions: Meng-Hui Wang conceived and designed the experiments; Mei-Ling Huang wrote thepapper; Zi-Yi Zhan analyzed the data; Chong-Jie Huang performed the experiments.

Conflicts of Interest: The authors declare no conflict of interest.

References

1. Karki, R.; Billinton, R. Considering renewable energy in small isolated power system expansion.In Proceedings of the 2003 Canadian Conference on Electrical and Computer Engineering, Montreal, QC,Canada, 4–7 May 2003; pp. 367–370.

2. Ai, B.; Yang, H.; Shen, H.; Liao, X. Computer-aided design of PV/wind hybrid system. IEEE Photovolt.Energy Convers. 2003, 3, 2411–2414. [CrossRef]

3. Chen, H.C.; Bai, W.X. Dynamic modeling and simulation of renewable energy based hybrid power generationsystems. In Proceedings of the 8th Intelligent Living Technology Conference, Taichung, Taiwan, 7 June 2013;pp. 1048–1055.

4. Hiyama, T.; Kouzuma, S.; Imakubo, T.; Ortmeyer, T.H. Evaluation of neural network based real timemaximum power tracking controller for PV system. IEEE Trans. Energy Conv. 1995, 10, 543–548. [CrossRef]

5. Li, F.T.; Chao, Q.; Dai, X.J. The research of wind power optimal capacity configuration in hydraulicpower system. In Proceedings of the Third International Conference on Electric Utility Deregulationand Restructuring and Power Technologies, Nanjing, China, 6–9 April 2008; pp. 2569–2574.

6. Chandel, S.S.; Ramasamy, P.; Murthy, K.S.R. Wind power potential evaluation of 12 locations in westernHimalayan region of India. Renew. Sustain. Energy Rev. 2014, 39, 530–545.

7. Nelson, D.B.; Nehrir, M.H.; Wang, C. Unit sizing and cost analysis of stand-alone hybrid wind/PV/fuel cellpower generation systems. Renew Energy 2006, 31, 1641–1656. [CrossRef]

8. Ekren, O.; Ekren, B.Y. Size optimization of a PV/wind hybrid energy conversion system with battery storageusing simulated annealing. Appl. Energy 2010, 87, 592–598. [CrossRef]

9. Yang, H.X.; Zhou, W.; Lu, L.; Fang, Z.H. Optimal sizing method for stand-alone hybrid solar-wind systemwith LPSP technology by using genetic algorithm. Sol. Energy 2008, 82, 354–367. [CrossRef]

10. HOMER ENERGY. Available online: http://www.homerenergy.com (accessed on 4 March 2016).11. IHoGA (Improved Hybrid Optimization by Genetic Algorithms). Available online: http://personal.unizar.es/

rdufo/index.php?option=com_weblinks&view=category&id=9&Itemid=119&lang=en (accessed on 7March 2016).

12. Taguchi, G.; Elsayed, E.A.; Hsiang, T.C. Quality Engineering in Production Systems; McGraw-Hill Book Co.:New York, NY, USA, 1989; pp. 1–10.

13. Small wind turbines, 1–6 kW. Greenspec. Available online: http://www.greenspec.co.uk/building-design/small-wind-turbines/ (accessed on 4 March 2016).

14. Borowy, B.S.; Salameh, Z.M. Optimum Photovoltaic Array Size for a Hybrid Wind/PV System. IEEE Trans.Energy Conv. 1994, 9, 482–488. [CrossRef]

Energies 2016, 9, 174 17 of 17

15. Steiner, N.Y.; Mocoteguya, P.; Candussoc, D.; Hissel, D.; Hernandez, A.; Aslanides, A. A review on PEMvoltage degradation associated with water management: Impacts, influent factors and characterization.J. Power Sources 2008, 183, 260–274. [CrossRef]

16. Lingfeng, W.; Singh, C. Compromise between cost and reliability in optimum design of an autonomoushybrid power system using mixed-integer PSO algorithm. In Proceedings of the International Conference onClean Electrical Power, Capri, Italy, 21–23 May 2007; pp. 682–689.

17. Cai, W. The extension set and incompatibility problem. J. Sci. Explor. 1983, 1, 81–93.18. Dingwall, R.B. Development and product integration of a reliable optoelectronic system using Taguchi

method. In Proceedings of the Annual Reliability and Maintainability Symposium, Orlando, FL, USA, 29–31January 1991; pp. 541–546.

19. Huang, K.H.; Shu, W.C. Multi-objective optimized design of permanent magnet motor by using the taguchimethod with robust rules. In Proceedings of the 9th Intelligent Living Technology Conference (ILT2014),Taichung, Taiwan, 6 June 2014; pp. 202–205.

20. Wang, M.H.; Chao, K.H. Application of the Extension Neural Network-Type 3 to Defect Recognition of CarEngine. ICIC Express Lett. 2012, 6, 417–423.

21. Wang, M.H.; Huang, M.L.; Liou, K.J. Application of Extension Theory with Chaotic Signal Synchronizationto Detect Islanding Effect of Photovoltaic Power System. Int. J. Photoenergy 2015, 2015. [CrossRef]

22. Wang, M.H.; Huang, M.L.; Liou, K.J. A Novel Islanding Detection Method for Grid Connected PhotovoltaicSystems. IET Renew. Power Gener. 2015, 6, 700–709. [CrossRef]

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open accessarticle distributed under the terms and conditions of the Creative Commons by Attribution(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).